v0.17.0

What's Changed

• Test the install.sh with shellcheck by @aliasaria in #303
• Part 1: Added Exporter Logging Functionality by @aahaanmaini in #287
• Part 2: Added Exporter Logging Functionality by @aahaanmaini in #288
• Added ByteDance Seed Coder Model by @Sourenm in #307
• Fix samsum dataset path by @mina-parham in #308
• Adding support for AMD ROCm by @deep1401 in #259
• Small changes to not break AMD when building from source on a dev machine by @deep1401 in #313

New Contributors

• @mina-parham made their first contribution in #308

Full Changelog: v0.16.2...v0.17.0

v0.16.2

hotfix

v0.16.1

What's Changed

• test workflow db by @aliasaria in #284
• Switch from nvidia-ml-py3 to nvidia-ml-py by @deep1401 in #285
• Fix broken pytest import by @deep1401 in #291
• Added the ability to pull model group gallery from galleries on startup by @Sourenm in #293
• Install.sh uses reqs file from RUN_DIR if available otherwise fallback to TLAB_CODE_DIR by @deep1401 in #290
• Fix single file names being fetched from .tlab_markitdown by @deep1401 in #292
• Update dataset gallery to match galleries one by @deep1401 in #295
• Remove Unused Requirements from all requirements files by @deep1401 in #298
• Update to local model-group-gallery.json by @Sourenm in #301

New Contributors

• @Sourenm made their first contribution in #293

Full Changelog: v0.16.0...v0.16.1

v0.16.0

What's Changed

• Upgrade wandb to avoid all protobuf 6.xx errors by @deep1401 in #271
• Adding MLX PPO Trainer by @deep1401 in #269
• add our first basic tests for the api by @aliasaria in #272
• Set manual context length for fallback on MLX Server by @deep1401 in #273
• Move API tests to test client instead of running a live server by @deep1401 in #275
• Rewrote exporter plugins to use plugin SDK by @aahaanmaini in #274
• Initial changes to support CUDA 12.8 (required for NVIDIA 50-series GPUs) by @aliasaria in #263
• Fix pytests installing dependencies by @deep1401 in #278
• Fix -h flag in run.sh caused by changing from HOST to TLABHOST by @deep1401 in #279
• Add back tests for install and server run by @deep1401 in #280
• Add option to load models in 4-bit for unsloth bnb models by @deep1401 in #282
• Fix failing imports on exporter plugins by @deep1401 in #276
• Fix updating names of tasks by @deep1401 in #283

New Contributors

• @aahaanmaini made their first contribution in #274

Full Changelog: v0.15.3...v0.16.0

v0.15.3

What's Changed

• Fix install failure on cpu-only-devices by @deep1401 in #270

Full Changelog: v0.15.2...v0.15.3

v0.15.2

What's Changed

• Fix run async mode on LLM Judge and Red Teaming param for attack by @deep1401 in #260
• Change package from fschat to transformerlab-inference by @deep1401 in #265
• Rename data gallery to prevent local local init errors by @deep1401 in #262
• Pin DeepEval and add Langchain dependencies wherever necessary by @deep1401 in #264
• Plugin install API returns an error message on failure. by @dadmobile in #261
• Separate builds for MacOS and CPU-only devices by @deep1401 in #267

Full Changelog: v0.15.1...v0.15.2

v0.15.1

What's Changed

• Move dataset gallery to be remote by @deep1401 in #256
• Add/allow manual install of plugin setup by @aliasaria in #257
• Move inference server model logs to a custom directory by @deep1401 in #258
• Fix RAG Reindexing and install the plugin if not already installed by @deep1401 in #250

Full Changelog: v0.15.0...v0.15.1

v0.15.0

What's New

• Comprehensive Qwen 3 Support across:
  • MLX Inference
  • MLX Training
  • MLX Export
  • CUDA Inference powered by Huggingface Transformers
  • SFT Training (single and multi GPU)
  • LoRA Training
  • GRPO Training (single and multi GPU)
  • Unsloth GRPO Trainer
  • DPO, ORPO, SIMPO Training
  • and more
• Custom dtypes for FastChat and vLLM allows for support for more graphics cards
• And more!

What's Changed

• Changes to FastChat and vLLM to work with custom dtypes by @deep1401 in #245
• MLX Server supports Gemma3 by @dadmobile in #243
• Add model delete from hf cache logic by @deep1401 in #249
• Upgrade MLX-LM to 22.1 by @aliasaria in #251
• Add Qwen3 to the model gallery by @aliasaria in #252
• Upgrade MLX and add support for Qwen3 in Trainer and Exporter by @deep1401 in #254
• Add Qwen3 Support to GGUF Exporter. by @aliasaria in #255
• Add Qwen3 support to LlamaTrainer plugin by @aliasaria in #253
• Llama CPP Upgrade by @deep1401 in #248

Full Changelog: v0.14.1...v0.15.0

v0.14.1

Hotfix problem with mamba create command

v0.14.0

What's Changed

• Change from Miniconda to Miniforge by @brunodoamaral in #203
• Upgrade transformers to 4.51.3 by @deep1401 in #229
• Fix sys.executable executions within plugins by @deep1401 in #227
• Fixed: GPU did not access : GPU shows NaN by @kolithawarnakulasooriya in #232
• Add support for models stored using XET storage on HF Hub by @deep1401 in #233
• Upgrade Pydantic things from .dict() to .model_dump() by @deep1401 in #234
• refactor workflows runner by @sanjaycal in #236

New Contributors

• @brunodoamaral made their first contribution in #203
• @kolithawarnakulasooriya made their first contribution in #232

Full Changelog: v0.13.5...v0.14.0